Microsoft blocks Fable access for employees

Welcome back, Superhuman. Anthropic's newest model has proven both powerful and controversial. Impressive demos of Fable's capabilities have flooded the timeline, but its built-in guardrails have locked certain groups out — sparking backlash over who should control access to powerful models.

Today: Google’s experimental new model, how to dictate anywhere on your Android device with AI, and get the latest prompts and trending social posts.

1. Microsoft limits employees' use of Fable over security and privacy concerns: The tech leader has removed Fable 5 from its internal system, due to the model’s data retention rules: storing conversations for 30 days and flagged content for up to two years. Fable has also drawn criticism for its safety classifiers, which prevent prompts on biology, chemistry, cybersecurity, and model distillation.

2. Dario Amodei calls for binding regulation of frontier AI models: In a rare personal post, Anthropic’s CEO published Policy on the AI Exponential. The essay argues that AI progress is outpacing political frameworks and it’s time to move to mandatory third-party testing of frontier models. He outlines five societal policy areas that need reimagining in an AI world. Read the full essay.

3. Google open-sources DiffusionGemma for up to 4X faster text generation: The experimental new model takes an entirely different approach to text generation. Instead of predicting one word at a time like most modern LLMs, it generates 256 tokens in parallel, producing up to 4X faster inference on dedicated GPUs. The speed gains could be a significant breakthrough for AI applications that depend on real-time results. Learn more here.

Turn every customer touchpoint into your competitive advantage. Get agents working on every account: run discovery, track deals, gather product feedback, and handle what used to take days.

Whether you're working in your browser, inbox, or favorite AI tool, connect to your customer data in real-time through Attio’s web app, MCP, API, and SDK.

You deserve a CRM that matches your ambition. That’s Attio.

Start now (at no cost)

Paying the bill. The days of using AI to rewrite the same email five different times may be coming to an end. And quickly. OpenAI CEO Sam Altman said some customers now view AI budgeting as a “huge issue,” something that hasn’t come up until this year. The real problem isn’t overspending on AI — it’s spending blindly, with little insight into how AI is translating to productivity or profit.

Blind spending. Many companies spent 2025 urging employees to experiment with AI, but they may have gotten carried away. Top firms now spend $7,500 per employee per month on AI, with OpenClaw creator Peter Steinberger blowing through $1.3M worth of tokens in a single month by himself. The message for 2025 was to experiment freely. In mid-2026, companies are tightening the purse strings.

Getting selective. One way to cut AI spend dramatically? Model routing, or using cheaper models for easier tasks. For example, building an internal software tool likely requires the most powerful model available. But drafting a weekly report? That can be done on an older one. Companies can build routing logic into workflows, or just teach employees how to use AI more efficiently.

Going on-device. Dr. Olena Zhu, Head of AI Solutions at Intel, argues that on-device AI models can be used alongside cloud-based frontier models to save on costs, noting that “while everyone chases larger cloud models, the real moat is the on-device control layer.” The AI spending spree isn't ending — it’s just facing more scrutiny.

Available on Mac, Windows, iPhone, and Android. Free and unlimited on Android during launch.

Download Wispr Flow and turn your thoughts into text instantly.

Most decision-makers don’t need any more proof that AI agents matter. They need a better way to decide what to build, how to measure it, and what to scale. Learn how to evaluate agents beyond demos, time-saved claims, and vendor noise.

Download the Guide.

📚 Bookmark Worthy: Your old workflows probably won’t work as well on Claude Fable. Here are four strategies to use it more effectively, directly from an Anthropic researcher.

🖌 CanvaGPT: You can now edit your ChatGPT-generated images directly inside Canva, the popular design app — giving you more editing precision with AI images. The announcement post racked up 3M views.

💻 Fable Projects: Explore 8 of the wildest projects people have created using Anthropic’s most powerful model, including a Minecraft clone and an original piece of music.

📝 Design Glossary: Want better results from your AI? Try using terms from this viral design glossary to craft more precise prompts.

🤯 Stock Returns: These 5 AI companies have two things in common: their stocks are up over 100% in the past year, and their HQs all fit into the same Google Maps screenshot.

^{* indicates a promoted tool, if any}

Readers loved this Best Practices Guide for Prompting Claude Fable 5, posted by Anthropic’s team to help you get the most out of its powerful new model.

Grow customers & revenue: Join companies like Amazon, HubSpot, and Salesforce. Showcase your product to our 1M+ readers and 2M+ followers on socials. Get in touch.

Until next time — Zain, Theodore, & the Superhuman AI team.